CARLOS GUESTRIN
Assistant Professor, Machine Learning, , Computer Science
www

My research interests are in sensor networks, artificial intelligence and machine learning , in particular, tackling large-scale stochastic dynamic systems by designing theoretically-founded algorithms unifying methods from statistics, optimization theory and machine learning, and by exploiting problem-specific structure. The goal is then to use these methods to address issues that arise in real-world problems, typified by wireless sensor networks, including robustness to failures and scaling to very large domains.

Our current projects seek, in the long-run, to develop the required algorithms and methods to scale up decision-making and inference methods to large-scale distributed systems where the state of the world is not fully observable. Specific topics include:

Distributed inference in sensor networks: algorithms for coherent probabilistic inference in sensor networks while minimizing communication cost, and robustly responding to network failures. Inference (or state estimation) is an indispensable building block for decision-making tasks. Our efficient methods are leveraged by representation and inference approaches from probabilistic graphical models, and by novel networking and message passing algorithms.

Efficient distributed multiagent coordination, planning and learning: building on the combination of probabilistic graphical models with Markov decision processes, we design efficient approximate approaches that enable distributed planning and learning in collaborative multiagent settings, where multiple decision makers must coordinate their actions to maximize a common goal.

Cost sensitive query optimization in distributed systems: in sensor networks, the energy cost of acquiring a measurement can vary widely for different types of sensors, and these measurements can be highly correlated. We design algorithms for obtaining query plans that minimize the expected cost of answering a query, including the cost of collecting data from the network, by exploiting these correlations between attributes.

Modeling and learning with temporally and spatially-correlated data: we design models and algorithms for learning stochastic temporal structure in sensor data, taking into account short-term correlations and long-term cyclical trends. When combined with efficient planning techniques, such temporal models can allow us to solve decision-making problems in real-world deployments. In addition, we develop efficient learning algorithm that combine the margin maximization and high-dimensional features (kernels) from support vector machines (SVMs), and the ability to exploit problem structure from graphical models, to obtain strong theoretical generalization bounds and empirical performance. When combined with our distributed inference techniques, this method can provide an effective solution for distributed classification in sensor networks.

For more information and latest publications, please visit http://www.cs.cmu.edu/~guestrin.